U.S. patent number 5,987,460 [Application Number 08/888,017] was granted by the patent office on 1999-11-16 for document retrieval-assisting method and system for the same and document retrieval service using the same with document frequency and term frequency.
This patent grant is currently assigned to Hitachi, Ltd.. Invention is credited to Yoshiki Niwa, Hirofumi Sakurai.
United States Patent |
5,987,460 |
Niwa , et al. |
November 16, 1999 |
Document retrieval-assisting method and system for the same and
document retrieval service using the same with document frequency
and term frequency
Abstract
Because the whole image of a retrieved document group cannot be
viewed, next retrieval request after one retrieval attempt has been
determined only perceptually. Therefore, displaying a means for
displaying topic words on a display means, extracting a group of
words characteristically appearing in a retrieved document group on
request from a user, further examining the relation between topic
words, and preparing a graph using the topic words as nodes, the
whole image of the retrieval results is displayed on the means for
displaying topic words. Additionally, by selecting an interesting
word or a word with no interest on a displayed graph of topic
words, the user can design a subsequent retrieval strategy
effectively.
Inventors: |
Niwa; Yoshiki (Saitama-ken,
JP), Sakurai; Hirofumi (Saitama-ken, JP) |
Assignee: |
Hitachi, Ltd. (Tokyo,
JP)
|
Family
ID: |
16008964 |
Appl.
No.: |
08/888,017 |
Filed: |
July 3, 1997 |
Foreign Application Priority Data
|
|
|
|
|
Jul 5, 1996 [JP] |
|
|
8-176174 |
|
Current U.S.
Class: |
1/1; 707/E17.082;
707/E17.058; 707/999.006; 707/999.004; 707/999.005 |
Current CPC
Class: |
G06F
16/338 (20190101); Y10S 707/99936 (20130101); Y10S
707/99934 (20130101); Y10S 707/99935 (20130101) |
Current International
Class: |
G06F
17/30 (20060101); G06F 017/30 () |
Field of
Search: |
;707/5,2,6,3,4,10 |
References Cited
[Referenced By]
U.S. Patent Documents
Other References
Proceedings of the 1st ACM International Conference on Digital
Libraries Interactive Term Suggestion for Users of Digital
Libraries: Using Subject. .
Thessauri and Co-Occurence Lists for information Retrieval, Schatz
et al. pp. 126-133 Mar. 20, 1996. .
Ablex, Article 12, 1990, "Information Retrieval Using Pathfinder
Networks", Fowler et al., pp. 165-179. .
Proceedings of ACM Sigir '92, "Scatter/Gather: A Cluster-based
Approach to Browsing Large Document Collections", Cutting et al,
pp. 318-329..
|
Primary Examiner: Black; Thomas G.
Assistant Examiner: Rones; Charles L.
Attorney, Agent or Firm: Fay, Sharpe, Beall, Fagan, Minnich
& McKee
Claims
What is claimed is:
1. A method for assisting document retrieval, comprising,
detecting documents each including one or a plurality of keywords
corresponding to a query as documents of retrieval results from a
retrieval subject document group,
detecting the document frequency of a word, representing the number
of documents where the word appears in the group of the documents
of retrieval results,
detecting the total document frequency of a word representing the
number of documents where the word appears in the whole retrieval
subject document group,
introducing a frequency ratio representing the ratio of the
document frequency of a word to the total document frequency of the
word,
classifying the document frequency in a set of frequency classes as
based on a given relation and assigning each word to a
corresponding frequency class, depending on the document frequency
of the word,
extracting an appropriate number of words from each of the
frequency classes in the decreasing order of frequency ratio of
word as topic words, and
displaying the extracted topic words in the form of a graph or a
list.
2. A method for assisting document retrieval, according to claim 1,
wherein the extracted topic words are displayed in the form of a
list per frequency class or the form of a graph representing the
relation between topic words.
3. A method for document retrieval service, comprising,
detecting documents each including a keyword transmitted from a
retriever as documents of retrieval results from a retrieval
subject document group,
detecting the document frequency of a word, representing the number
of documents where the word appears in the group of documents as
retrieval results,
detecting the total document frequency of a word representing the
number of documents where the word appears in the whole retrieval
subject document group,
introducing a frequency ratio representing the ratio of the
document frequency of a word to the total document frequency of the
word,
classifying the document frequency in a set of frequency classes
based on a given relation and assigning each word to a
corresponding frequency class, depending on the document frequency
of the word,
extracting an appropriate number of words from each of the
frequency classes in the decreasing order of frequency ratio of
word as topic words,
composing the extracted topic words as a data displayable in the
form of a list per frequency class or in the form of a graph
representing the relation between topic words, and
transmitting the topic words and said composed data for displaying
to the retriever.
4. A method for document retrieval service, according to claim 3,
wherein the retriever is equipped with a means for transmitting
keywords for retrieval to the retrieval system, and receiving the
composed data concerning topic words transmitted from the retrieval
system, and displaying the topic words according to the transmitted
composed data, so as to receive retrieval service.
5. A method for document retrieval service, according to claim 4,
wherein the retriever receives, from a document retrieval service
system before acting document retrieval for transmitting keywords
for retrieval to the retrieval service system, and receiving the
composed data concerning topic words transmitted from the retrieval
service system, and displaying the topic words according to the
transmitted composed data.
6. A machine readable data storing media on which the word
frequency data for selecting topic words are recorded, wherein the
frequency data of each word comprises (a) character sequence, (b)
the document frequency of the word, representing the number of
documents where the word appears in the group of documents of
retrieved results, (c) the total document frequency of the word
representing the number of documents where the word appears in the
whole retrieval subject document group, (d) the frequency ratio
representing the ratio of the document frequency of the word to the
total document frequency of the word, and (e) the frequency class
of the word assigned to the word depending on its document
frequency, and wherein topic words are extracted from each of the
frequency classes in the decreasing order of frequency ratio of
word.
7. A machine readable data storing media on which the co-occurrence
data for calculating the relatedness among topic words are
recorded, wherein the co-occurrence data of each pair of topic
words comprises (a) the co-occurrence frequency of the word pair,
that is, the number of documents in the retrieved document set
where both words of the word pair appear, and (b) the co-occurrence
intensity of the word pair, such as their co-occurrence frequency
divided by document frequency of the second word of the word pair,
and wherein the links of the graphical display of topic words are
generated for word pairs with strong relation.
8. A machine readable data storing media on which the coordinate
data for displaying a graphical display of topic words are
recorded, wherein the data comprises data for displaying nodes of
the topic word graph and data for displaying links of the graph
representing strong relation between topic words, and the data for
displaying nodes comprises the coordinate center, character
sequence, and the character number in the crosswise and lengthwise
directions of a region displaying the characters, and the size of
the displaying region, and the data for displaying links comprises
the initiation coordinate and the termination coordinate of each
link, and wherein the graphic display of a word graph are displayed
following the data.
9. A machine readable data storing media on which are recorded the
data of each word for selecting topic words, data of each pair of
topic words for calculating the relatedness among topic words, and
data of each topic word for displaying a graphical display of topic
words, wherein the data of each word comprises (a) character
sequence, (b) the document frequency of the word, representing the
number of documents where the word appears in the group of
documents of retrieved results, (c) the total document frequency of
the word representing the number of documents where the word
appears in the whole retrieval subject document group, (d) the
frequency ratio representing the ratio of the document frequency of
the word to the total document frequency of the word, and (e) the
frequency class of the word assigned to the word depending on its
document frequency, and topic words are extracted from each of the
frequency classes in the decreasing order of frequency ratio of
word, wherein the co-occurrence data of each pair of topic words
comprises(a) the co-occurrence frequency of the word pair, that is,
the number of documents in the retrieved document set where both
words of the word pair appear, and (b) the co-occurrence intensity
of the word pair, such as their co-occurrence frequency divided by
document frequency of the second word of the word pair, and the
links of the graphical display of topic words are generated for
word pairs with strong relation, wherein the data comprises data
for displaying nodes of the topic word graph and data for
displaying links of the graph representing strong relation between
topic words, and the data for displaying nodes comprises the
coordinate center, character sequence, and the character number in
the crosswise and lengthwise directions of a region displaying the
characters, and the size of the displaying region, and the data for
displaying links comprises the initiation coordinate and the
termination coordinate of each link, and the graphic display of a
word graph can be displayed following the data.
10. A document retrieval system comprising
a means for detecting documents each including one or a plurality
of keywords corresponding to a query as documents of retrieval
results from a retrieval subject document group,
a means for detecting the document frequency of a word,
representing the number of documents where the word appears in the
group of the documents of retrieval results,
a means for detecting the total document frequency of a word
representing the number of documents where the word appears in the
whole retrieval subject document group,
a means for introducing a frequency ratio representing the ratio of
the document frequency of a word to the total document frequency of
the word,
a means for classifying the document frequency in a set of
frequency classes as based on a given relation and assigning each
word to a corresponding frequency class, depending on the document
frequency of the word,
a means for extracting an appropriate number of words from each of
the frequency classes in the decreasing order of frequency ratio of
word as topic words, and
a means for displaying the extracted topic words in the form of a
graph or a list.
11. A document retrieval system, according to claim 10, wherein
said display means displays the extracted topic words in the form
of a list per frequency class or in the form of a graph
representing the relation between topic words and said retrieval
system includes means for selecting the display form.
12. A document retrieval system, according to claim 11, wherein the
query comprises three types of keywords of requisite keyword,
positive keyword and negative keyword, and the retrieval by
requisite keyword is carried out under AND conditions with each
requisite keyword, and wherein individual documents in the group of
documents as retrieval results can be given a higher score
depending on the number of positive keywords if the documents
include the positive keywords or are reduced of their scores
depending on the number of negative keywords if the documents
include the negative keywords, to extract topic words from a
document group with a higher score given.
13. A document retrieval system, according to claim 12, wherein the
type of a keyword can be changed among three types of keywords,
namely the one or a plurality of keywords of the query, positive
keywords and negative keywords, and the topic words displayed can
be copied as any of these three types of keywords.
14. A document retrieval system, according to claim 12, wherein
retrieval with positive keywords is run if no requisite keyword is
set and retrieval with each positive keyword is run under OR
conditions.
15. A document retrieval system, according to claim 14, wherein the
type of a keyword can be changed among three types of keywords,
namely the one or a plurality of keywords of the query, positive
keywords and negative keywords, and the topic words displayed can
be copied as any of these three types of keywords.
16. A document retrieval system, according to claim 11, wherein the
relation between topic words is determined on the basis of the
co-occurrence relation between the topic words, and wherein the
form of a graph is a graph composed by linking together a word pair
of topic words with higher relation, using topic words as
nodes.
17. A document retrieval system, according to claim 16, wherein the
direction of the longitudinal axis in the graphic representation of
topic words represents the document frequency of a topic word in
the retrieved document group.
18. A document retrieval system, according to claim 11, wherein the
direction of the longitudinal axis in the graphic representation of
topic words represents the document frequency of a topic word in
the retrieved document group.
19. A document retrieval system, according to claim 3, wherein the
relation between topic words is determined on the basis of the
co-occurrence relation between the topic words, and wherein the
form of a graph is a graph composed by linking together a word pair
of topic words with higher relation, using topic words as
nodes.
20. A document retrieval system, according to claim 19, wherein the
direction of the longitudinal axis in the graphic representation of
topic words represents the document frequency of a topic word in
the retrieved document group.
21. A document retrieval system, according to claim 10, wherein the
query comprises three types of keywords of requisite keyword,
positive keyword and negative keyword, and the retrieval by
requisite keyword is carried out under AND conditions with each
requisite keyword, and wherein individual documents in the group of
documents as retrieval results can be given a higher score
depending on the number of positive keywords if the documents
include the positive keywords or are reduced of their scores
depending on the number of negative keywords if the documents
include the negative keywords, to extract topic words from a
document group with a higher score given.
22. A document retrieval system, according to claim 21, wherein the
type of a keyword can be changed among three types of keywords,
namely the one or a plurality of keywords of the query, positive
keywords and negative keywords, and the topic words displayed can
be copied as any of these three types of keywords.
23. A document retrieval system, according to claim 21, wherein
retrieval with positive keywords is run if no requisite keyword is
set and retrieval with each positive keyword is run under OR
conditions.
24. A document retrieval system, according to claim 23, wherein the
type of a keyword can be changed among three types of keywords,
namely the one or a plurality of keywords of the query, positive
keywords and negative keywords, and the topic words displayed can
be copied as any of these three types of keywords.
Description
BACKGROUND OF THE INVENTION
The present invention relates to a document retrieval-assisting
method having a user interface to attain an interactive guidance
function for document retrieval and a system for the same and a
document retrieval service using the same.
For document retrieval, a variety of interfaces between document
retrieval systems and users have been designed and developed, so
that these users can readily reach a desired document assembly.
Feedback and guidance are primary interfaces among them. Feedback
is a mechanism that when a user draws his judgment of YES/NO
concerning several items as the results of retrieval, retrieval
results reflecting the judgment can then be gained. Further,
guidance is a function to provide information with relation to a
retrieval operation at the individual processes of the retrieval
operation, namely information believed to possibly work as a
reference when a user intends to modify or improve the retrieval
conditions.
As to the guidance function, conventionally, a method has generally
been conducted, comprising proposing information relating to the
input retrieval conditions. For example, a method is illustrated,
comprising storing a database representing relations between words,
such as thesaurus, and retrieving from the data base a set of words
with relation to the input word as one retrieval condition.
Thesaurus is a tree-structure database primarily showing the is-a
relations between words, however, a method is also suggested,
comprising automatically generating a data of related words using
co-occurrence statistics [see for example B. R. Schatz et al.,
Interactive term suggestion for users of digital libraries: Using
subject thesauri and co-occurrence lists for information retrieval.
Proc. ACM DL '96. P.126-133]. A method is additionally proposed,
comprising displaying a retrieving word and words with relation to
the word in a network structure using the co-occurrence statistic
data between words [see for example, R. H. Fowler, D. W. Dearholt,
Information Retrieval Using Pathfinder Networks. In Pathfinder
Associative Networks, Ablex, article 12, Edited by R. W.
Schvaneveldt (1990)].
However, the method providing information with relation to a
retrieving condition is disadvantageous in that the method is
hardly applicable to a case with a plurality of query words or a
case with negation used; and the method is also problematic in that
the method is hardly applicable to the document retrieval with no
use of any keyword (such as associative search). So as to overcome
these problems, a method is proposed, comprising automatically
extracting related information from retrieval results and providing
the information to a user. For example, Scatter/Gather method [D.
Cutting, et al. (1992), Scatter/Gather: A Cluster-based Approach to
Browsing Large Document Collections. Proc. ACM SIGIR '92, p.
318-329] automatically classifies a retrieved document group
(clustering) and displays the topic words therein per each class.
However, real time response is hardly effected in the case of
clustering, because the increase of the number of documents
escalates the number of calculations by an order of second and
third powers; as the progress in retrieval operation, generally,
the difference between classes becomes so delicate that the
characteristic properties of a class can hardly be hinted from the
topic words of the class.
SUMMARY OF THE INVENTION
So as to overcome the aforementioned problems and propose a set of
topics contained in a retrieved document group on real time, it is
an object of the present invention to display the topic words
characteristically appearing in a document group in a graph form or
a list form; it is another object of the present invention to
provide a document retrieval assisting-method capable of a well
balanced extraction of low-frequency words to high-frequency words
characteristically appearing in the document group and a system
therefor; and it is an additional object of the present invention
that a user intending to use the document retrieval system (method)
can carry out document retrieval at a remote distance.
In order to display topic groups contained in a retrieved document
group on real time, therefore, word groups characteristically
appearing in a document group are defined as nodes; when an intense
co-occurrence relation is observed between a pair of topic words,
namely when the number of documents containing both of the words is
very large, the word pair is linked to compose a graph which is
displayed, and for graphic representation of the topic words, the
document frequency of the topic words is represented on the
longitudinal axis so as to identify general words and words with
high specificity at glance. As to an example of the display of
topic words in a list form, topic words are firstly divided in
frequency classes, and words at high document frequencies middle
frequencies and low frequencies should be aligned in the separate
list boxes.
So as to extract topic words with good balance from words at low
frequencies and words at high frequencies, in selecting topic words
in a retrieved document group, the topic words are classified by
occurrence frequency. Then, from the individual classes, topic
words should be extracted in the decreasing order of the ratio of
the document frequency in the retrieved document group to the
document frequency in the entire document database.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a block diagram depicting the composition example of a
retrieval system comprising computers for independent use in an
embodiment of the present invention;
FIG. 2 is a view depicting one arrangement example of assigning
data in work areas;
FIG. 3 is a view depicting one example of the display image of the
user-interface window for information retrieval;
FIG. 4 is a view depicting one example of the data stored in the
work area for searching during retrieval operation;
FIG. 5 is a view depicting one example of the retrieval results
displayed on the display image of the user-interface window for
information retrieval as shown in FIG. 3, after the retrieval
operation;
FIG. 6 is a view depicting one example of the display image when
the window for displaying topic words is started, through which a
user can provide a topic word as a keyword for searching;
FIG. 7 is a view depicting one example of the data stored in the
area for storing a graph of topic words on request of a user to
display topic words;
FIG. 8 is a view depicting one example of graphic representation of
topic words extracted from the retrieved document group;
FIG. 9 is a view depicting one example of a word frequency data of
the retrieved document group;
FIG. 10 is a view depicting one example of a list of topic words
extracted from the retrieved document group;
FIG. 11 is a view depicting one example of a data representing
co-occurrence relation between topic words generated from the
retrieved document group;
FIG. 12 is a view depicting one example of a list of pairs of topic
words with an intense co-occurrence relation in the retrieved
document group;
FIG. 13 is a PAD (Problem Analysis Diagram) view depicting one
example of the composition of a calculation routine calculating the
graphic arrangement of topic words;
FIG. 14 is a PAD view depicting one example of a method for
calculating x-coordinate in graphic arrangement;
FIG. 15 is a view depicting one example of the coordinate data in
virtually arranging the graphic representation of the retrieval
results in a normalized region;
FIG. 16 is a view depicting one example of the coordinate data of a
graph for graphic representation of the retrieval results, before
the overlapping of the displayed nodes in a graph is solved;
FIG. 17 is a PAD view depicting one example of the detail of the
routine to solve the overlapping of displayed nodes in a graph;
FIG. 18 is a view depicting one example of graphic representation
of topic words, when the number of displaying topic words is set at
20;
FIG. 19 is a view depicting one example of the display image of the
user-interface window for information retrieval, being equipped
with a means for selecting a topic word display mode;
FIG. 20 is a view depicting one example of the display of topic
words in list form; and
FIG. 21A is a block diagram representing the composition example of
the side of a server, and FIG. 21B is a block diagram representing
the composition example of the side of a client, in a case such
that the principal body of a retrieval system is provided to the
side of a server and a plurality of clients make access to the
system for carrying out retrieval.
DESCRIPTION OF THE PREFERRED EMBODIMENTS
EMBODIMENT I
The embodiment I of the present invention will now be described
below with reference to FIGS. 1 to 20. The present embodiment is
one composition example of a retrieval system comprising computers
for independent use. In the present embodiment, retrieval results
represented in graph form are primarily described.
FIG. 1 shows the overall composition of the document retrieval
system of the present embodiment; 1 represents input device; 2
represents display device; 3 represents CPU; 4 represents storage
device for computing program; 5 represents work area for operating
the computing program; and 6 represents data storage device. These
means or devices are in communication by means of bus 100 to
exchange signals between them.
The input device 1 comprises keyboard 11, mouse 12 and optical
pen-type input device 13. On the display device 2 are displayed
user-interface window for information retrieval 21 and window for
displaying topic words 22, so as to guide retrieval. The storage
device for computing program 4 contains a routine for managing
user-interface of information retrieval 41, morphological analysis
routine 42, retrieval routine 43 and topic words display routine
44, which are essential for the document retrieval system of the
present embodiment. In order to extract and display topic words
from a retrieved document group on the window for displaying topic
words 22, the topic words display routine 44 uses topic words
extracting routine 441, co-occurrence analysis routine 442, graph
mapping routine 443 and graph displaying routine 444, as the
sub-routines. The detail of the work area 5 is described below with
reference to FIG. 2. The data storage device 6 comprises text
database 61, index files for retrieval 62, word frequency database
63 and stop word database 64. As these databases, generally, a user
selects those satisfying the retrieval object of herself as
retrieval data subjects, and then uses them. For example, a user
may purchase and use a database issued from a newspaper company
when she wants to search newspaper articles. It is needless to say
that stop word database 64 may sometimes be offered as an attached
data from the supplier of the system.
FIG. 2 shows the detail of the composition of the work area 5. The
work area 5 is a work area for storing parameters or transient
data, required for the operation of various routines at the storage
device for computing program 4, and comprises work area for
searching 51, work area for extracting topic words 52, work area
for co-occurrence analysis 53 and work area for graph mapping 54.
The individual areas are equipped with further finely divided data
areas, but the detail thereof will be described when the routines
individually relating to the areas are started.
When a user wishes document retrieval, the command to start the
document retrieval system is input from the keyboard 11.
Correspondingly, the routine for managing user-interface of
information retrieval 41 is initiated, to display the
user-interface window for displaying information retrieval 21 to
progress retrieval operations in an interactive manner on the
display device 2.
FIG. 3 is one example of the initial image of the user-interface
window for displaying information retrieval 21. The user-interface
window for displaying information retrieval 21 comprises area for
entering query 211, area for displaying and controlling keywords
212, area for displaying the number of retrieved documents 213,
area for displaying title list 214, area for displaying text 215,
button for search 216 and button for displaying topic words
217.
In the present embodiment, as the keywords for document retrieval,
use is made of three types of keywords, namely requisite keyword,
positive keyword, and negative keyword. Retrieval is carried out by
AND of requisite keyword(s); when no requisite keyword is given, OR
of positive keyword(s)is used. Each document is given a score which
is the number of positive keywords contained in the document minus
the number of negative keywords contained in the document. Even if
an identical keyword appears repeatedly in one document, one point
is simply added or subtracted. Corresponding to these three types
of keywords, area for displaying and controlling keywords 212 are
composed of three parts. The compositions in these three parts are
identical, so the left part for requisite keywords are mainly
described herein. The part of the requisite keywords on the area
for displaying and controlling keywords 212 comprises area of
requisite keywords 2121, move button 21211 and clear button 21212.
The move button 21211 is used to transfer another type of keywords
to the part of requisite keywords, while the clear button 21212 is
used to eliminate the keywords from the part of requisite keywords.
More specifically, if a keyword displayed on the part of positive
keywords or negative keywords is selected and the move button 21211
for requisite keywords is pushed, the selected keyword is
transferred to the part of requisite keywords. If a keyword
displayed on the part of requisite keywords is selected and the
clear button 21212 is pushed, the selected keyword is eliminated
from the part of requisite keywords. If a keyword displayed on the
part of requisite keywords is selected and the move button 21221
for positive keywords is pushed, the selected keyword is
transferred to the part of positive keywords. If a keyword
displayed on the part of requisite keywords is selected and the
move button 21231 for negative keywords is pushed, the selected
keyword is transferred to the part of negative keywords.
As will be described below, additionally, these move buttons can be
used as copying buttons to copy the topic words on display of topic
words. In other words, the use of the button, namely the use for
moving or the use for copying, is determined, depending on where a
subject word is located.
For inputting retrieval requirement, window for entering query 2111
of the area for entering query 211 is clicked by mouse 12 to put
the area at a state for waiting for an input, and using the
keyboard 11, then, the retrieval requirement such as requisite
keywords, positive keywords and negative keywords should be input.
When pushing the button for converting query to keywords 2112,
subsequently, the character row input in the window for entering
query 2111 is transferred on morphological analysis routine 42
where the character row is divided in word rows. Then, with
reference to the stop word database 64, the words registered
therein are eliminated, and the results are stored in the area for
storing keywords of default type 5111 or 5112 of the area for
storing keywords 511 (FIG. 2). Here we assume the default keyword
type is negative. The individual contents are displayed in the form
of a list on areas for storing keywords 2121 or 2122. As apparently
shown in an example below, in this case, the state of word division
is determined on the basis of the dictionary of the morphological
analysis routine 42.
Pushing button for search 216, then, retrieval routine 43 is
started, and with reference to the index files for retrieval 62
(data linking each word to the set of document representing which
document containing the word), documents including all of requisite
keywords are retrieved. The row of identification numbers of
retrieved documents is stored in area for storing search results
512. The retrieval routine 43 carries out such an operation that
one point is added per one positive keyword contained in the
document retrieved with the requisite keywords, while one point is
subtracted per one negative keyword contained therein. The scores
are also stored, together with the document identification numbers,
in the area for storing search results 512. With no assignment of
any requisite keyword, the retrieval routine 43 carries the
retrieval by OR of positive keywords; and in the same manner, the
scores are calculated hereinafter. With no requisite keyword or no
positive keyword, no retrieval is run even if the button for search
216 is pushed.
Because requisite keywords are processed with AND for retrieval,
the keywords are inevitable for more strictly narrowing the search
results. When wishing search with retrieval results rather with no
leakage, on contrast, the search should be run with OR, using only
positive keywords; when an item possibly including an undesirable
item in the retrieval results may be assumed, negative keywords
should be set.
The retrieval routine 43 furthermore calculates the score
distribution on the basis of the retrieval results stored in the
area for storing search results 512, and the results are then
stored in the area for storing score distribution 513. The term
score distribution means a data showing the number of documents
with each score of positive points or negative points.
Description will be made of an example wherein retrieval
requirement by using "electronic publishing" as a requisite keyword
is input.
After inputting the character row of "electronic publishing" as the
requisite keyword on the window for entering query 2111, the button
for converting query to keywords 2112 is pushed. Through the
morphological analysis routine 42, the "electronic publishing" is
divided as "electronic/publishing" and stored in the area for
storing requisite keywords 5111 and further displayed as divided on
the first line and the second line of the area of requisite
keywords 2121.
FIG. 4 shows the state of work area for searching 51 when the
button for search 216 is pushed at that stage. In the present
example, the requisite keywords are "electronic" and "publishing",
and therefore, they are stored in the area for storing requisite
keywords 5111. Besides them, any positive keyword or negative
keyword is never provided by a retriever on the window for entering
query 2111, and therefore, the area for storing positive keywords
5112 and the area for storing negative keywords 5113 are left as
blank. The retrieved document numbers and the scores are stored in
the area for storing search results 512. In this case, because of
the absence of positive keywords or negative keywords, all the
scores are zero. The data of the number of retrieved documents per
each score are stored in the area for storing search results 513.
In this case, the score is simply zero, with 77 cases.
FIG. 5 depicts the state of the user-interface window for
information retrieval 21, displaying the search results. The
requisite keywords, "electronic" and "publishing", are displayed on
the area of requisite keywords 2121; the area for displaying the
number of retrieved documents 213 displays the contents of the area
for storing score distribution 513; and the area for displaying
title list 214 displays an appropriate number of the retrieved
document identification numbers and the titles on one line per one
case. When the document identification number and the title, not
yet displayed, are wanted, the part appearing on the display should
be scrolled by means of a so-called scroll bar. If you want to read
the contents from any displayed title, the corresponding title is
pointed with a mouse and the like, to display a part of the
contents on the area for displaying text 215. If you want to know
the contents of a document not on display, a part appearing on the
display should be scrolled by means of a scroll bar, in the same
fashion.
Above all, totally 77 documents are retrieved, which are related to
"electronic publishing". Retrieval is sometimes narrowed to a
specific subject at next stage; otherwise, an overview of what
types of topics may be contained in the 77 documents is wanted. In
such case, pushing the button for displaying topic words 217 on the
user-interface window for information retrieval 21 (FIG. 3), topic
words display routine 44 is started to display window for
displaying topic words 22 on the display device 2.
FIG. 6 depicts the detail of a part of the window for displaying
topic words 22. The window for displaying topic words 22 is
composed of operation area 221, area for displaying and controlling
keywords 222, area for displaying the number of retrieved documents
223, area for displaying topic words 224 and area for setting
parameters 225. The area for displaying and controlling keywords
222 and the area for displaying the number of retrieved documents
223 are individually in communication with the area for displaying
and controlling keywords 212 and the area for displaying the number
of retrieved documents 213 of the user-interface window for
information retrieval 21, respectively, and when the contents
displayed thereon are changed by the operation on the window for
displaying topic words 22, the individual displays of the
user-interface window for information retrieval 21 automatically
change as well. Change in the adverse direction, namely the change
of keywords and the number of retrieved documents by the operation
on the user-interface window for information retrieval 21, cannot
be reflected automatically on the window for displaying topic words
22. So as to incorporate the change, the contents of the side of
the user-interface window for information retrieval 21 are copied
on the side of the window for displaying topic words 22 when the
reset button 2214 on the operation area 221 is pushed. On the
initial display of the window for displaying topic words 22
displayed on the display device 2, by pushing the button for
displaying topic words 217 on the user-interface window for
information retrieval 21, the keywords and the number of retrieved
documents on the user-interface window for information retrieval 21
are automatically copied. In the present example, "electronic" and
"publishing" are displayed on the area of requisite keywords 2221,
while on the area for displaying the number of retrieved documents
223 is displayed "score 0: 77 cases".
Pushing then the button for displaying topic words 2212 of the
operation area 221, topic words extraction routine 441 is started,
to read the document identification numbers with the maximum score
from the data stored in the area for storing search results 512 and
then analyze the contents of the documents corresponding to the
document identification numbers to display words (topic words)
characteristically contained in the documents as well as the
results in a graph of the analysis of the relation of the topic
words on the area for displaying topic words 224. The process will
be described in detail in the following description.
FIG. 7 is an example of "electronic publishing", representing the
data stored in area for storing a graph 543 (FIG. 2). The graph is
composed of nodes and links, which are stored in area for storing
nodes 5431 and area for storing links 5432, respectively. The nodes
data to be stored comprises the topic word (character row)
represented by each node, the coordinate center of the topic word
represented in the area for displaying topic words 224, the
character numbers in the crosswise and lengthwise directions of a
region displaying the characters, and the size of the displaying
region (provided that the size of the displaying region is
represented by its half value, namely the size from the center to
the ends for ready use). Alternatively, the link data to be stored
is composed of the initiation coordinate and the termination
coordinate of a line to be represented on a graph. In the figure,
in addition to the initiation coordinate and the termination
coordinate to be stored in the area for storing links 5432, the
data of the character rows corresponding to individual links are
shown for reference. But the data are not required for an actual
system.
FIG. 8 is a view depicting the window for displaying topic words 22
at a state such that a graph of topic words is displayed by pushing
the button for displaying topic words 2212 on the operation area
221. Following the data of the area for storing a graph 543, the
graph displaying routine 444 represents a graph composed of topic
words and links connecting the words, on the area for displaying
topic words 224. From the data in FIG. 7, for example, "compact" is
displayed on a rectangular display region with the coordinate
center present on the coordinate (149, 131) and with the number of
characters in the crosswise direction being 3 and the line number
being 2, which region is of a size of 27 from the center to both
the sides in the crosswise direction and of 18 from the center to
the upper and lower sides in the lengthwise direction. In the
present example, the crosswise direction is defined as rightward
direction and the lengthwise direction is defined as downward
direction, provided that the coordinate is defined as the upper
left of the area for displaying topic words 224. Additionally, the
link data can be defined as the coordinates of the initiation point
and the termination point. The first one of the link data means the
connection between the coordinate centers of the topic words
"publishing" and "electronic", while the second data means the line
from the coordinate (203, 131) to the coordinate (308, 40). For
representation of these words, an opaque rectangle is displayed on
the background for character display in the display region of each
node, to hide the line in the node display region, thereby
preparing a readily viewable graph. When the graphic line
representing a link overlaps with the nodes display region,
however, no line appears, which may induce erroneous judgment. In
the data in FIG. 7, for example, the graphic line connecting
between "desk top publishing" and "a publication" passes through
the display region of "Nifty Serve", and therefore, the graphic
line does not appear as a line in the region if "Nifty Serve" is
displayed on an opaque rectangle background. Consequently, it looks
as if "Nifty Serve" and "a publication" were connected together
with the graphic line and additionally as if "Nifty Serve" and
"desk top publishing" were connected together with the graphic
line. As a countermeasure against such phenomenon, in FIG. 8, such
a display is adopted that the display of a graphic line is omitted
around the initiation point and termination point of the graphic
line, instead of the display of an opaque rectangle, so that it is
avoided for the graphic line to enter into the nodes display
regions, and additionally that the passing of the graphic line is
apparently viewable in other display regions. It is very difficult
to make arrangement so as to avoid the disappearance of graphic
lines in an opaque rectangle if attached, and particularly, if it
is intended to display a great number of topic words, a display at
a readily viewable size might never been attained eventually.
The device for setting the number of topic words to be displayed
2251 of the area for setting parameters 225 is for adjusting the
number of words to be displayed on the area for displaying topic
words 224. By sifting the button for setting 22511 to left or
right, the device 2251 can be set to a desirable numerical figure.
The set value is displayed on display device 22512, and is stored
in area for storing the number of topic words to be extracted 5213
of area for storing parameters for topic words extraction 521. The
numerical figure is utilized by the topic words extraction routine
441.
The process of preparing a graphic data as shown in FIG. 7, after
the button for displaying topic words 2212 of the window for
displaying topic words 22 is pushed, will be described below. When
the button for displaying topic words 2212 is pushed, the topic
words extraction routine 441, co-occurrence analysis routine 442
and graph mapping routine 443, stored in storage device for
computing programs 4, are sequentially initiated.
The topic words extraction routine 441 reads the maximum score and
the number of retrieved documents from area for storing search
distribution 513 of the work area for searching 51. In the example
of "electronic" and "publishing", the maximum score (S) is 0, while
the number is 77. Then, maximum documents number for word frequency
counting (M) 5211 is read out from the area for storing parameters
for topic words extraction 521. (Herein, M=300.) Because a longer
time is needed to analyze all the documents of which the retrieved
document number K is large and if the number K exceeds a given
limit for M, samples of the number M should be extracted.
With reference to the area for storing search results 512, the
topic words extraction routine 441 reads the contents of all the
document identification numbers with the scores agreeing with the
maximum score (S) from the text database 61, and using the
morphological analysis routine 42, the contents are divided in
words to count the number of documents where each of all the words
appears (referred to as document frequency hereinbelow). In this
example, the number of the retrieved documents with the maximum
score is 77 and the maximum documents number for word frequency
counting is M=300 or less. Therefore, all the documents are
read.
Furthermore, the morphological analysis of a subject document can
be run, by storing the results of the morphological analysis of all
the documents and reading the results if the database storage means
has enough capacity. Because of no need of practicing the
morphological analysis per retrieval, then, the analysis time can
be markedly shortened effectively.
The data of words and document frequencies thereof, thus recovered,
are stored in the area for storing frequency data 523 in the work
area for extracting topic words 52. Because the results of the
morphological analysis of the subject document as described above
are to be used subsequently, the results are stored in the area for
storing documents segmented into words 522.
FIG. 9 shows one example of "electronic publishing", wherein a part
of the data stored in the area for storing frequency data 523 is
shown. Each data is composed of five items, namely the name of a
word, the document frequency, the total document frequency, the
frequency ratio and the frequency class. The document frequency of
a word is the number of retrieved documents where the word appears.
Furthermore, the total document frequency is the number of
documents using the word in the entire retrieval document subjects,
with no relation with the results of retrieval. From the
information stored in word frequency database 63 is extracted the
frequency information of a subject word. Thus, the word frequency
database 63 should be prepared beforehand, by preliminarily
scanning the entire documents as retrieval subjects to count all
the words appearing in the documents and calculate the total
document frequencies of the words. The frequency ratio is a value
of the document frequency divided by the total document frequency.
For a case of first word ROM, for example, the frequency ratio is
21 divided by 1183 being equal to about 0.017, provided that 21 is
the document frequency and 1183 is the total document
frequency.
Then, the frequency class is described below. A topic word
characteristic to a document group can be determined on the basis
of the dimension of the frequency ratio, and a larger frequency
ratio means a higher degree of topic properties. However, it is
risky to compare two words with significantly different document
frequencies from each other. A word with a lower frequency
potentially has a higher probability of a larger frequency ratio
because such word has a lower total frequency. For example, in FIG.
9, the frequency ratio of "desk top publishing" is 0.75 but the
word cannot be defined as a word with greater topic properties due
to the larger frequency ratio. This is because the document
frequency is only 3, while the total document frequency is also
only 4. So as to avoid comparison of words with larger difference
in the document frequency, therefore, the document frequency is
preliminarily divided broadly at an appropriate interval for
classification, and in each class, words with larger frequency
ratios can be selected as topic words, whereby topic words can be
extracted with good balance from words at a lower frequency to
words at a higher frequency.
One example as to how to determine the frequency class is described
below. Topic words extraction routine 441 reads number of frequency
classes (C) 5212 as a parameter as to how many frequency classes
should be used for division, which is determined by a user. Herein,
C=5 (generally, C is an integer of 1 or more). Provided that the
i-th frequency ratio is C[i], the document frequency required to be
classified in the C[i] is defined as f[i] or more to less than
f[i+1]. However, the largest class has f[i+1] or less instead of
less than f[i+1]. It will be described then how to determine the
value of f[i] as the frequency threshold value. Herein, provided
that K' is the subject document number, f[i]=K' in the order of
(i/(C+1)) power. (When the number of the retrieved documents does
not exceeds the maximum documents number for word frequency
counting M, K'=K; if K>M, K'=M.) In this example, K'=77 and C=5,
and therefore, f[1]=77 in the order of (1/6) power=2.06, f[2]=4.25,
f[3]=8.77, f[4]=18.10, and f[5]=37.33. Thus, class 1 is defined as
a class of words with a document frequency of 3 or more to 4 or
less; class 2, of 5 or more to 8 or less; class 3, of 9 or more to
18 or less; class 4, of 19 or more to 37 or less; and class 5, of
38 or more to 77 or less.
According to the classification provisions, the frequency class of
a word can be determined on the basis of the document frequency of
the word. In the case of ROM, the document frequency is 21, so the
word is classified in class 4; additionally, "interactive" has a
document frequency of 5, and is therefore classified in class 2. If
the document frequency is smaller than that for class 1 (in this
case, the document frequency is 2 or less), the word is removed
from the topic words extraction subjects.
The frequency class can be directly calculated by the following
formula and is then given to a word. But if the value is equal to
C, 1 is subtracted from the value.
Frequency class=[a maximum integer not above
Subsequently, the topic words extraction routine reads number of
topic words to be extracted (p) 5213 and extracts topic words of
the number in total, sequentially from those ranked in the
decreasing order of the frequency ratio in each frequency class. As
one example of the method for running the extraction, a method is
illustrated, comprising extracting topic words of the number (n+1)
from the classes with a frequency class of 1 or more to r or less
and extracting topic words of the number n from the classes with a
frequency class of r or more, provided that the number of extracted
words p divided by the number of division of frequency classes C is
the quotient n with the remainder r.
The example shown in FIG. 9 is described, provided that the number
of extraction p is 10. Because the number of division C is 5, the
quotient n of p divided by C is 2 with the remainder r being 0.
Thus, two topic words are uniformly extracted from each of the
classes 1 to 5. From the data in the area for storing frequency
data 523, two topic words are extracted sequentially in the
decreasing order of the frequency ratio from each of the individual
frequency classes. If the words in the class 5 are to be aligned in
the decreasing order of the frequency ratio from the data of FIG.
9, the row of "publishing" (0.027), "electronic" (0.015), "media"
(0.006), and "information" (0.001) is produced. Thus, the two topic
words at the upper rank, namely "publishing" and "electronic", can
be extracted. In the same manner, hereinbelow, ROM and "compact"
are extracted as the topic words from the class 4; "mail" and "a
publication", from the class 3; "interactive" and "Nifty Serve",
from the class 2; and "desk top publishing" and "publishing", from
the class 1. These are stored in the area for storing topic words
524.
FIG. 10 shows examples of the data stored in the area for storing
topic words 524. Topic words extracted by the above process are
stored, together with their document frequencies. In the figure,
the frequency classes are also shown for reference, but they can be
omitted satisfactorily.
By the above process, the topic words extraction routine 441 is
passed through. Continuously, co-occurrence analysis routine 442
analyzes the co-occurrence relation between topic words, and the
results are stored in the area for storing co-occurrence data
531.
The area for storing co-occurrence data 531 is of a two-dimensional
alignment, containing in its lengthwise and crosswise directions
the topic words assembly stored in the area for storing topic words
524. The individual elements represent the number of documents in
which a corresponding word pair appears commonly. The co-occurrence
analysis routine 442 reads the retrieved document group segmented
in words, from the area for storing documents segmented into words
522, and increments an element in the area for storing
co-occurrence data 531 for all pairs of topic words co-occurring in
each document.
Then, the co-occurrence analysis routine 442 calculates the
co-occurrence intensity of each pair of topic words. The
co-occurrence intensity is a value of the co-occurrence frequency,
which is firstly counted by the above procedure and then divided by
the document frequency as the latter word of a word pair
(corresponding to the word in the row, in the table). As the
document frequency of word, the value stored in the area for
storing topic words 524 is used (FIG. 10).
FIG. 11 represents the data stored in the area for storing
co-occurrence data 531 at this stage. Each grid is composed of two
numerical figures, wherein the upper figure shows the co-occurrence
frequency of a corresponding word pair, while the lower figure
shows the co-occurrence intensity of the word pair (co-occurrence
frequency divided by document frequency of a word on the side of
row). For example, the numerical FIG. 6 on the upper part of the
grid on line 6 and row 3 means that the topic word "a publication"
on line 6 and the topic word ROM at row 3 occur concurrently in 6
documents. In this case, because the document frequency of the
topic word of the word pair on the row side, namely ROM, is 21, the
numerical figure of the co-occurrence intensity on the lower part
is about 0.29 as the value of 6 divided by 21. In the area for
storing co-occurrence data 531, the topic words are arranged in the
decreasing order of document frequency. Because the lower half of
the table below the diagonal line is used in the subsequent
process, the remaining part is omitted herein.
Continuously, the co-occurrence analysis routine 442 extracts word
pairs with larger co-occurrence intensities (pairs to be linked
together on the graph of topic words), from the co-occurrence data.
In the present example, each topic word X is to be linked to a
topic word with the maximum value of the co-occurrence intensity
with X, among words with higher document frequencies than X. The
co-occurrence analysis routine 442 collects word pairs to be linked
together according to this standard, which are then stored in the
area for storing co-occurrence links 532.
When a word with the second largest co-occurrence intensity or the
third largest co-occurrence intensity does not have such a markedly
small co-occurrence intensity, compared with the word with the
maximum co-occurrence intensity (for example, a word with an
intensity 0.9-fold or more the maximum intensity), linking to the
word is also powerful.
FIG. 12 is a view depicting the contents of the area for storing
co-occurrence links 532 at its stage. The process of extracting
these links is described with reference to the example of FIG. 11.
As to the word "publishing" in the second line in FIG. 12, a single
word with a larger document frequency than the frequency of
"publishing" is "electronic", and therefore, "publishing" is linked
to "electronic". As to the word ROM on the third line, then, two
words of "publishing" and "electronic" have higher frequencies than
the frequency of ROM, and the co-occurrence intensities thereof
with ROM are both 0.27. In this case, linking is effected to
"publishing" with a smaller number in the area for storing
co-occurrence data 531. Then, the fourth word "compact" is
examined. The co-occurrence intensity of the word "compact" with
the third word ROM is the largest and is 0.81. Thus, from
"compact", linking is effected to ROM. The same procedures are
continued below, to recover a link data as shown in FIG. 12.
By the above procedures, the co-occurrence analysis routine 442 is
passed through to initiate graph displaying routine 443. Based on
the data in the area for storing topic words 524 (FIG. 10) and the
data in the area for storing co-occurrence links 532 (FIG. 12), the
procedure to arrange a graph of topic words on a two-dimensional
plane is conducted.
FIG. 13 shows the detail of the graph mapping routine 443. The
graph mapping routine 443 is composed of routine for calculating y
coordinate 4431, routine for calculating x coordinate 4432, routine
for converting these coordinates into display coordinate 4433,
routine for solving overlap of nodes 4434 and link mapping routine
4435, and these are initiated in this order.
On the assumption that the display region is a square display
region of [-1, 1]'[-1, 1], routine for calculating y coordinate
4431 and the routine for calculating x coordinate 4432 calculate
coordinates to map individual nodes. The coordinates are called
normalized coordinates. The calculated coordinates data is stored
in the area for storing normalized coordinate 541.
Firstly, the routine for calculating y coordinate 4431 is
initiated. According to the calculation formula
y=(6/.pi.).times.arctan [0.2.times.log(f/fm)), a normalized y
coordinate of a position to display each topic word is calculated
on the basis of the document frequency thereof. That is, a topic
word with a larger document frequency is arranged on an upper part
on the y coordinate. Herein, fm represents the frequency of a topic
word located in the center of topic words aligned in the decreasing
order of document frequency [under the provision that the topic
word to be selected is located at (the number of total topic words
divided by 2)+1, if the number is even]. In the example, the
document frequency of "electronic" "publishing" is 77 and is
located at the uppermost part, while the document frequency of
"publication" is 9 and is located in the center position. .pi.
means the ratio of the circumference of a circle to its diameter;
logarithm log means natural logarithm; arctan means the reciprocal
function of tangent function; angle is represented in unit radian.
For example, the frequency of "compact" is 21 and therefore, its
normalized y coordinate is about 0.32 by the formula
(6/.pi.).times.arctan (0.2.times.log(21.div.9)). Normalized y
coordinates of other topic words are calculated in the same
manner.
Then, the routine for calculating x coordinate 4432 is initiated to
calculate a normalized x coordinate of a position to display each
topic word.
FIG. 14 is a view depicting the detail of the routine for
calculating x coordinate 4432. Firstly, nodes never having parent
nodes (linking nodes) are collected by step 44321. In this case,
only the node of "electronic" is such node. Thus, the value of the
x coordinate is calculated by the formula xi=-1+2i/(r+1) in the
step 44321, provided that i=1, as follows;
-1+(2.times.1)/(1+1)=0.
Then, enter loop 44322. At step 44323, a node linked to the node
with the defined x coordinate (in this case, "electronic") is
singly selected. From the data of co-occurrence links (FIG. 12),
herein, it is indicated that "publishing" satisfies the
conditions.
Continuously, enter step 44324. Determining the assembly of parent
nodes of the node selected at the step 44323, the average of the x
coordinates thereof is calculated. The assembly of the parent nodes
of "publishing" is {"electronic"}, and the average of the x
coordinates is 0.
Then at step 44325, nodes of which the assembly of the parent nodes
agrees with {"electronic"} are collected. Herein, such node is
singly "publishing".
Continuously, enter branched step 44326. Because the average of the
x coordinates of the parent nodes is 0, step 44327 is selected to
calculate the x coordinate of "publishing". Inserting s=1, xp=0 and
i=1 in the calculation formula of the step 44327, the x coordinate
of "publishing" is calculated as 0.
By the above procedures, the normalized x coordinates of
"electronic" and "publishing" are determined. Because the x
coordinates of all the nodes are not yet determined, loop 44322 is
repeated. Among the nodes with still undermined x coordinates at
the step 44323, one of nodes without any link except those to
"electronic" and "publishing" is selected. In this case, ROM
satisfies the conditions.
At step 44324, an assembly of those to which ROM is linked is
determined, to give {"publishing"}. Additionally, the average xp of
the x coordinates of the parent node {"publishing"} is calculated
as 0.
At step 44325, nodes which are linked to the assembly corresponding
to {"publishing"} are to be collected. Other than ROM, "mail"
satisfies the conditions. Because the average xp of the x
coordinates of the parent nodes is 0, the upper branch 44326 is
selected. At step 44327, the x coordinates of ROM and "mail" are
individually calculated as -0.33, and 0.33, by equally dividing
[-1, 1] by three. By the same manner, nodes being linked to only
the nodes with determined x coordinates and having the common
linking assembly are collected, and then, their x coordinates are
determined in order that the nodes are uniformly arranged within
the interval (-1, 1] while the average of the x coordinates of the
parents is in the center.
FIG. 15 is an example of "electronic publishing", representing the
coordinate data stored in area for storing the normalized
coordinate 541 at this stage.
Continuously, the graph mapping routine 443 starts the routine for
conversion into display coordinates 4433, whereby the coordinates
normalized in the region [-1, 1]'[-1, 1] are converted into
coordinates representing the practical position on the area for
displaying topic words 224 and are then stored in the center
coordinate column of the area for storing nodes 5431 (FIG. 16).
Conversion is carried out by the following primary formula.
Herein, x and y in small characters are normalized coordinates,
while X and Y in large characters are the coordinates on the area
for displaying topic words 224. ym represents the maximum value of
y. In the example of FIG. 15, ym=0.774. As coefficients R.sub.x,
R.sub.y, O.sub.x and O.sub.y, values stored in the subject area of
the area for storing parameters for graph mapping 542 (FIG. 2) are
used, respectively. In the present example, R.sub.x =200, R.sub.y
=200, O.sub.x =60 and O.sub.y =40.
For the case of "compact", the normalized coordinates by the
primary conversion are (-0.555, 0.320), and calculation is
conducted as follows;
By such manner, the actual coordinates of all the nodes on the area
for displaying topic words 224 are calculated and stored in the
area for storing nodes 5431 (FIG. 16). As the preparation for the
next step, then, words are aligned in the increasing order of x
coordinates. Calculating the number of characters h in the cross
wise direction and line number v and calculating the crosswise size
H and lengthwise size V of the word display region, these values
are stored in the area for storing nodes 5431.
The size of the word display region is calculated by the following
formula. Characters should be written in the crosswise direction,
with the crosswise size limit of words being defined as W. As the
value of W, a value stored in the maximum number of characters in a
line for displaying a topic word 5426 is used. Herein, W=3. When
the number of characters to be displayed is M, the number of
characters in the crosswise direction h and the line number v are M
and 1, respectively, under the provision of M.ltoreq.W. If M>W,
h equals to W, while v is the minimum integer above the value of M
divided by W. As to "electronic", for example, the number of
characters is 2 below W=3 as the crosswise width limit, therefor
the line number v=1 and the cross wise size h=2. Next, as to
"interactive", the number of characters is 8, which is over W=3,
the line number v is the minimum integer above 8/3, namely 3, with
the crosswise width h being equal to W=3. Furthermore, half size H
of the crosswise size and half size W of the lengthwise size are
individually calculated from the number of characters h and v,
respectively, by the following formula. The reason why such half
size values are set resides in that the half size values are
primarily used at the following processes.
Herein, F represents the size of character font; m.sub.x represents
the dimension of the margin along the x direction; and m.sub.y
represents the dimension of the margin along the y direction. So as
to avoid too close alignment of two nodes, m.sub.x and m.sub.y
represent the essentially minimum interval between two nodes. As
the F, m.sub.x and m.sub.y, values stored in the font size 5425,
and horizontal margin for displaying topic words 5427 and vertical
margin for displaying topic words 5428 (FIG. 2), respectively, are
used. In the present example, F=16, m.sub.x =3 and m.sub.y =2. For
example, in the case of "compact", h=3 and v=2, and therefore, the
following calculation is done.
The number of characters and the size of display area in the area
for storing nodes 5431 are thus calculated.
By such manner, coordinates on the area for displaying topic words
are determined, but at this stage, a possibility of nodes
overlapping remains. In the example of FIG. 16, for example, the
coordinate of "electronic" are the same as that of "publishing",
and therefore, they are overlapped to each other. Thus, routine for
solving overlap of nodes 4434 is started, to run the procedure to
sift the coordinates so as to solve overlapping.
FIG. 17 shows the detail of the routine for solving overlap of
nodes 4434.
All nodes are sorted in the increasing order of the x coordinates,
which are defined as N[1], . . . , N[r]. The coordinate of N[i] is
defined as (X[i], Y[i]), while the value of the character display
region size is (H[i], V[i]). The following procedures are done
provided that i=2, . . . , r. As to j in [1 i-1] satisfying
.vertline.Y[j]-Y(i).vertline.<V[i]+V[j], the maximum value of
X[j]+H[j] is defined as .epsilon.. When no such j is found, the
procedure to sift the coordinate concerning this i is not
necessary. .delta.=.epsilon.-(X[i]-H[i]). If .delta..ltoreq.0, the
procedure to sift the coordinate as to i is not necessary. If
.delta.>0, overlapping occurs and therefore, all of the x
coordinates of N[i] . . . , N[r] are sift by d to right. In other
words,
By the above process, coordinates to display all nodes without
overlap can be given.
As to "interactive" at i=2, the data of FIG. 16 shows that
Therefore, the formula .vertline.Y[2]-Y[1].vertline.<V[2]+V[1]
cannot be established. Thus, no procedure for horizontal sifting is
run as to "interactive".
Then, the case of i=3, namely ROM, is examined. If j=1,
.vertline.Y[3]-Y[1].vertline.=.vertline.131-131.vertline.=0,
V[3]+V[1]=10+18=28. Then,
.vertline.Y[3]-Y[1].vertline.<V[1]+V[3]. That is, overlapping
with "compact" at j=1 occurs. As to the relation with "interactive"
at j=2,
.vertline.Y[3]-Y[2].vertline.=.vertline.131-240.vertline.=109,
V[3]+V[2]=10+26=36. Therefore;
.vertline.Y[2]-Y[3].vertline.<.vertline.V[2]+V[3] is not
established. Thus, no possibility of overlapping with "interactive"
occurs. Therefore, only the case with j=1 should be considered
about the x coordinate. .epsilon.=X[1]+H[1]=/149+27=176. Hence, the
sift width d=.epsilon.-(X[i]-H[i])=176-(193-27)=10. Thus, X[j] is
absolutely +10 for any j from 3 to 10. Thus, (X[3], Y[3])=(203,
131). Hence, the coordinate of ROM can be produced as shown in FIG.
7. By repeating the same procedures thereafter, the same data as in
the area for storing nodes 5441 as shown in FIG. 7 can be
recovered. By the procedure to solve the overlapping of the
character display region, the overlapping of the character display
region with the graphic line cannot be checked. As an actual
problem, if it is intended to strictly solve the overlapping in the
limited display area, display in an appropriate size may sometimes
be impossible. Therefore in the present embodiment, it is intended
to conduct no check about this.
Finally, the graph mapping routine 443 starts link mapping routine
4435. The link mapping routine 4435 stores the information about a
word pair to be linked with co-occurrence, which is stored in the
area for storing co-occurrence links 532 in the work area for
co-occurrence analysis 53, and the routine 4435 prepares a line
data, namely the coordinates of an initiation point and a
termination point, to be displayed on the area for displaying topic
words 224 on the basis of the coordinate data of each node stored
in the area for storing nodes 5431 and then stores the line data in
the area for storing links 5422. For example, a link from ROM to
"publishing" is present in the area for storing co-occurrence links
532 of FIG. 12. From the data stored in the area for storing nodes
5431 of FIG. 7, it is apparently shown that the coordinate of ROM
is (203, 131) and the coordinate of "publishing" is (308, 40). The
data of a line from the initiation point of (203, 131) to the
termination point of (308, 40) is stored in the area for storing
links 5432. Thus, the data of a graph to be displayed is prepared
(FIG. 7).
An example of the applicable embodiment to progress a search
operation with reference to the graphic display of topic words
displayed on the area for displaying topic words 224 of the window
for displaying topic words 22 will be described hereinbelow.
FIG. 8 is an example of topic word display regarding "electronic
publishing", and herein, a user is virtually interested in one of
displayed words, for example "desk top publishing". In this case,
pointing the position of the word on display by means of mouse 12
and then pointing the move button 22222 for positive keywords,
"desk top publishing" is stored in the area for storing positive
keywords 5112, which is then displayed on the area of positive
keywords 2122 of the user-interface window for information
retrieval 21 and the area for positive keywords 2222 of the window
for displaying topic words 22. Subsequently, pushing the button for
search 216 of the user-interface window for displaying topic words
21 or search button 2211 of the window for displaying topic words
22, search is run while adding "desk top publishing" to the
positive keywords, to narrow the search.
When no interesting word is found among the topic words displayed
on the area for displaying topic words 224 of FIG. 8, the number of
displayed words can be increased by using device for setting the
number of topic words to be displayed 2251. FIG. 18 is an example
wherein the number of displayed topic words is increased to 20. In
this case, 20 words are selected from the data of FIG. 9, for
example, by topic word extraction routine 441, and in the same
manner as described in FIG. 8, these words are displayed. Herein,
if a user is virtually interested in "information retrieval" in the
field of electronic publishing, for example, the user can utilize
the "retrieval" and "information retrieval" displayed on the
displayed graph. Clicking "retrieval" and "information retrieval"
on the area for displaying topic words by means of mouse and
pushing move button 22222 toward positive keywords, these words are
added as positive keywords. Pushing thereafter search button 2211,
search can be narrowed. After narrowing search and for wishing to
see the graph of topic words for the narrowed result, button for
display of topic words 2212 is pushed, satisfactorily. If search
and graphic representation of topic words are continuously
conducted, then, only pushing of button for search+display of topic
words 2213 is satisfactory for effecting continuously the
aforementioned steps.
If "information retrieval" does not draw any attention or if the
documents regarding "information retrieval" are already checked and
topics other than the word draw attention, negative keywords are
utilized. If "retrieval" and "information retrieval" have been
added to positive keywords, these words are transferred from
positive keywords to negative keywords by pointing these words
displayed on the area of positive keywords 2222 by mouse and
pushing move button 22232 toward negative keywords. When intending
to directly use a word displayed on the area for displaying topic
words 224, the subject word is clicked with a mouse and the like to
push the move button 22232 toward negative keywords in the same
manner as for positive keywords. That is, in this example, the move
button conducts transfer procedure between retrieval keywords and
copying procedure from displayed topic words to keywords.
When search is practiced after transferring "retrieval" and
"information retrieval" to negative keywords, the score of a
document including these words is reduced while the score of a
document without these contained become relatively high.
Accordingly, a document with no relation to "information retrieval"
can be focused, among documents relating to "electronic
publishing".
FIG. 19 depicts one example of user-interface window for
information retrieval 21 equipped with means for selecting topic
word display mode 2171, having a function capable of displaying
topic words in the form of a graph or in the form of a list.
Compared with display in the form of a graph, the display in the
form of a list can display more topic words, and therefore, the
display cannot display the relation between topic words. Thus, the
results with attention focused on the relation among topic words
cannot be assessed, disadvantageously, but using a scroll bar, a
great number of topic words appearing in the retrieval results can
be listed, advantageously for a user to have a higher possibility
to find a related word satisfying the interest of the user.
Thus, selecting the graphical display mode by the means for
selecting topic word display mode 2171 shown in FIG. 19, firstly,
the search results are graphically shown to propose the overall
view of the whole image of topic words, including the individual
relations, and the results are assessed. When related words
satisfying the interest of a user never sufficiently appear in the
results, list display is used for further detailed search. Thus,
the results can be assessed at two stages. Furthermore, when an
interesting word is recovered from the display utilizing list, the
word can be used as a keyword to again start retrieval and repeat
the same procedure.
If graphic mode is selected by the means for selecting topic word
display mode 2171 in FIG. 19, graphic display of topic words can be
attained as explained in FIG. 8 or 18. If list mode is selected as
shown in FIG. 19, topic words can be displayed in the form of a
list on the area for displaying topic words 224, as shown in one
example in FIG. 20. Even if list is selected by the means for
selecting topic word display mode 2171, the method for extracting
topic words from the retrieved document group is the same as in the
case of graphic display as described above. For list display, the
frequency ranked in about 3 classes, namely high, middle and low
classes, is rather more viewable than the frequency ranked in about
5 classes as shown in FIG. 9, and therefore, the number of divided
frequency classes is 3 in the display example of FIG. 20.
In FIG. 20, corresponding to the selection of list, the
scroll-bar-equipped display frames of area for displaying
high-frequency topic words 2241, area for displaying
middle-frequency topic words 2242 and area for displaying
low-frequency topic words 2243 are set on the area for displaying
topic words 224, and topic words corresponding to the frequency
class data of the topic words in the area for storing frequency
data 523 are displayed in the individual display frames. The order
of display in each of the individual display frames is for example
the decreasing order of the frequency ratio, whereby a user can
visually look more general topic words than specific topic words
such as name and places. Thus, the user can search words satisfying
the interest from a wide variety of selective subjects.
EMBODIMENT II
The embodiment II of the present invention will be described below
with reference to FIG. 21. The present embodiment relates to a
search method, capable of satisfying retrieval requirement from a
plurality of users, while the embodiment I is a composition example
of a retrieval system comprising computers for independent use.
FIG. 21 shows an overall structure of another embodiment realizing
the document retrieval method of the present invention. In the
present embodiment, a plurality of clients make access through
signal transmission lines to a server, and search service can be
provided to each client. Generally, such server does not use the
method of itself as one client. However, in this embodiment, the
server is a retrieval system of substantially the same structure as
described in the embodiment I, plus communication means 7, taking
into account that the server can also need to utilize the method as
a client, in response to the remark of problems from the side of
clients. Client is composed of dashed means individually
corresponding to the input means 1, display means 2, CPU 3, storage
device for computing programs 4, work area 5 for operating the
computer program and bus 100, among the compositions described in
the embodiment I, and additionally composed of communication means
7' to make communication with server and printer 81 as output means
8. On the bus 100 of server is arranged interface IF1; on the bus
100' of client are arranged interface IF2 and IF3, and they are in
communication through lines NET1 and NET2 which connect between the
server and the client. As to the client 2, herein, only bus 100'
and interface IF2 are depicted, while others are omitted.
When client 1 intends to start document retrieval, firstly, the
command to start document retrieval system is input from keyboard
11' as the input means 1'. Correspondingly, the communication means
7' on the side of client is connected to the communication means 7
on the side of server through communication route NET1; The routine
for managing user-interface of information retrieval 41 of the
storage device for computing programs 4 on the side of server is
transmitted to the side of client 1, which is started on the side
of the client 1. Consequently, user-interface window for
information retrieval 21' to progress interactive retrieval
operation is displayed on the display device 2'. After displaying
of the user-interface window for information retrieval 21', the
client 1 utilizes this image to input a word functioning as a
retrieval key by the same procedures as described in the embodiment
I. Additionally, while retaining a copy of the routine for managing
user-interface of information retrieval 41 on the storage device
for computing programs 4', the client can start the copied routine,
satisfactorily.
It is convenient to receive the present search assisting service by
utilizing hyper text access interface such as WWW browser. In this
case, on the side of server is prepared hyper text (HT) for
transmitting the routine for managing user-interface of information
retrieval 41 to the side of client. Furthermore, it is a
preliminary provision that the side of client is in environment
such that common hyper text access interfaces can be utilized.
From the address input area of the hyper text access interface
displayed on the display device 2', assigning the address defined
by the present search assisting service (namely, the address of the
server on the net work and the file name in which the hyper text HT
to transmit the routine for managing user-interface of information
retrieval 41 is present), the hyper text HT is transmitted to the
side of client through the binary communication means, along with
the routine for managing user-interface of information retrieval
41, and the transmitted routine for managing user-interface of
information retrieval 41 is started by the computer on the side of
client, which displays the user-interface window for information
retrieval 21' on the display device 2' for use.
The address of the hyper text HT is designated directly above, but
when the address of the present hyper text HT is embedded as an
anchor in the hyper text displayed on the access device of the
hyper text access interface, the anchor part is clicked by mouse
and the like, whereby the same operation can be run.
The retrieval requirement input by the client 1 is transmitted
through the communication means 7, 7' and communication route NET1,
to the side of server, and the side of server practices necessary
retrieval, topic word extraction and computation of graphic
mapping, and the results are again transmitted back to the side of
the client 1 through the contact with communication means 7, 7',
which is then handed to the routine for managing user-interface of
information retrieval 41' of the client 1, where the routine
displays the topic word graph based on the data on the area for
displaying topic words 22'. As described in the embodiment I, on
the basis of the retrieval results, the client 1 further inputs a
data corresponding to further required retrieval procedures. The
data is then transmitted again to the side of server, where
necessary retrieval is practiced to display the results on the area
for displaying topic words 22'. The client 1 can utilize a
print-out output, printed by printer 81, if necessary.
The client 1 can utilize only the results practiced on the side of
server, with no waiting of substantial retrieval program. Thus, on
the side of the client 1, work area 5' requires only a capacity to
retain the initial input data and the retrieval results transmitted
from server and the topic words and the data relating to the
graphic arrangement, and thus, very rich retrieval service can be
proposed on a simple system.
* * * * *